In its simplest form, Regtech can be described as ‘the use of technology, particularly information technology, in the context of regulatory monitoring, reporting and compliance’ enabling it to provide ‘technological solutions to regulatory processes’. Regulatory technology (Regtech) has emerged mainly in financial services as a tool for regulatory compliance. However, to define Regtech as a sub-set of Fintech fails to properly explain its capabilities. Since Regtech has the potential for continuous monitoring capacity with close to real-time insights of domestic and global markets through artificial intelligence filters, it can be proactive rather than reactive, looking to identify problems in advance rather than after the fact. Research on Regtech and competition law has thus far considered how Regtech could be used by competition authorities to monitor compliance with relevant laws and regulations has been explored using statistical approaches (particularly in respect of cartels) and—to a more limited extent—using machine learning.
This article, available here, explores an approach to applying Regtech techniques to antitrust enforcement. It does this by applying those techniques to the detection of resale price maintenance (vertical price fixing). The article examines the application of machine learning in the Regtech environment and the ways in which application programming interfaces could be used.
Section II looks at prohibitions of resale price maintenance (RPM) clauses.
At its core, RPM refers to vertical restraints on price setting between a supplier and an acquirer of goods and services. The object of RPM prohibitions is to ensure that competition is unfettered by price restraints imposed by suppliers on re-suppliers of goods or services. In Australia, RPM is prohibited per se by section 48 of the Competition and Consumer Act 2010, but there is a mechanism whereby the ACCC can authorise the practice. In the EU, agreements that impose upon distributors and retailers a minimum or fixed resale price may be an infringement of Article 101 TFEU, but may be justified by efficiencies. In the US, RPMs used to be per se infringements of the Sherman Act, but are now subject to the rule of reason.
Section III considers RegTech technologies applicable to antitrust.
Machine learning has been identified as a potential basis for detection and enforcement in antitrust and particularly in cartel detection. Further, in some cases, a data interface to data could be used in order to conduct pricing analysis. This requires that the latest data is available from e.g. an e-commerce platform where trades are taking place. Using an API allows one to readily access the data underlying the application. If a data provider decides to allow an API, there is a choice between whether that interface is presented publicly or privately. An inefficient alternative is for data providers to engage in a process known as ‘screen scraping’ whereby one takes information from a web page designed for consumers to create a data set.
Section IV looks at algorithmic approaches to resale price maintenance.
This section provides an example of how RegTech could be used. It starts from the premise that the expected price for any particular good will be no higher than the recommended retail price and include some deviation below that price. This is verified by reference to the price of a well-known mobile phone. This exercise suggests that it would be feasible to collect a training data set for expected pricing to be used to identify potential infringements. In order to assess actual sale prices, the major e-commerce platforms provide APIs which can be used to monitor prices of specific goods. Then, relying on machine learning, the competition authority can build a library of expected price distributions against which to compare patterns found in the data. However, one specific form of machine learning, reinforced learning, will be of limited use given its requirement of numerous successful instances of detection.
The results of this exercise can be analysed by reference to some inferences regards RPM. Where an offered price is above the recommended retail price, this indicates overcharging. Where there is little deviation from the recommended retail price, this indicates that resale price maintenance may arise. Where pricing matches the expected distribution of prices, no further action would be required from the sample.
The adoption of such tools would need to overcome some technical challenges. First, there may be sampling issues e.g. arising from a reduced number of merchants. It is also possible that wholesalers will use different merchants on different e-commerce platforms, which can make comparisons more difficult unless all platforms are aggregated. However, even then merchants may be offering the product in question for different prices on different platform, e.g. reflecting different platform commissions. Such costs – including delivery charges – must be controlled for. In effect, such algorithmic control tools can overcome most such challenges, as long as it is recognised that they provide a screening tool that must be verified through further investigation.
This paper provides a nice examples of how ‘digital’ tools can be used to assist competition enforcement. It is rather interesting on the technical front, even if – as with the paper above – its claim to import a new approach to antitrust (in this case, from financial regulation) strikes me as a bit overdone. I am not familiar with machine learning screening tools for RPM, but then again I am not an expert but am nonetheless aware of numerous screening tools that rely on machine learning.